convert.c

   1 #include "cache.h"
   2 #include "attr.h"
   3 #include "run-command.h"
   4 #include "quote.h"
   5 #include "sigchain.h"
   6
   7 /*
   8  * convert.c - convert a file when checking it out and checking it in.
   9  *
  10  * This should use the pathname to decide on whether it wants to do some
  11  * more interesting conversions (automatic gzip/unzip, general format
  12  * conversions etc etc), but by default it just does automatic CRLF<->LF
  13  * translation when the "text" attribute or "auto_crlf" option is set.
  14  */
  15
  16 /* Stat bits: When BIN is set, the txt bits are unset */
  17 #define CONVERT_STAT_BITS_TXT_LF    0x1
  18 #define CONVERT_STAT_BITS_TXT_CRLF  0x2
  19 #define CONVERT_STAT_BITS_BIN       0x4
  20
  21 enum crlf_action {
  22         CRLF_UNDEFINED,
  23         CRLF_BINARY,
  24         CRLF_TEXT,
  25         CRLF_TEXT_INPUT,
  26         CRLF_TEXT_CRLF,
  27         CRLF_AUTO,
  28         CRLF_AUTO_INPUT,
  29         CRLF_AUTO_CRLF
  30 };
  31
  32 struct text_stat {
  33         /* NUL, CR, LF and CRLF counts */
  34         unsigned nul, lonecr, lonelf, crlf;
  35
  36         /* These are just approximations! */
  37         unsigned printable, nonprintable;
  38 };
  39
  40 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
  41 {
  42         unsigned long i;
  43
  44         memset(stats, 0, sizeof(*stats));
  45
  46         for (i = 0; i < size; i++) {
  47                 unsigned char c = buf[i];
  48                 if (c == '\r') {
  49                         if (i+1 < size && buf[i+1] == '\n') {
  50                                 stats->crlf++;
  51                                 i++;
  52                         } else
  53                                 stats->lonecr++;
  54                         continue;
  55                 }
  56                 if (c == '\n') {
  57                         stats->lonelf++;
  58                         continue;
  59                 }
  60                 if (c == 127)
  61                         /* DEL */
  62                         stats->nonprintable++;
  63                 else if (c < 32) {
  64                         switch (c) {
  65                                 /* BS, HT, ESC and FF */
  66                         case '\b': case '\t': case '\033': case '\014':
  67                                 stats->printable++;
  68                                 break;
  69                         case 0:
  70                                 stats->nul++;
  71                                 /* fall through */
  72                         default:
  73                                 stats->nonprintable++;
  74                         }
  75                 }
  76                 else
  77                         stats->printable++;
  78         }
  79
  80         /* If file ends with EOF then don't count this EOF as non-printable. */
  81         if (size >= 1 && buf[size-1] == '\032')
  82                 stats->nonprintable--;
  83 }
  84
  85 /*
  86  * The same heuristics as diff.c::mmfile_is_binary()
  87  * We treat files with bare CR as binary
  88  */
  89 static int convert_is_binary(unsigned long size, const struct text_stat *stats)
  90 {
  91         if (stats->lonecr)
  92                 return 1;
  93         if (stats->nul)
  94                 return 1;
  95         if ((stats->printable >> 7) < stats->nonprintable)
  96                 return 1;
  97         return 0;
  98 }
  99
 100 static unsigned int gather_convert_stats(const char *data, unsigned long size)
 101 {
 102         struct text_stat stats;
 103         int ret = 0;
 104         if (!data || !size)
 105                 return 0;
 106         gather_stats(data, size, &stats);
 107         if (convert_is_binary(size, &stats))
 108                 ret |= CONVERT_STAT_BITS_BIN;
 109         if (stats.crlf)
 110                 ret |= CONVERT_STAT_BITS_TXT_CRLF;
 111         if (stats.lonelf)
 112                 ret |=  CONVERT_STAT_BITS_TXT_LF;
 113
 114         return ret;
 115 }
 116
 117 static const char *gather_convert_stats_ascii(const char *data, unsigned long size)
 118 {
 119         unsigned int convert_stats = gather_convert_stats(data, size);
 120
 121         if (convert_stats & CONVERT_STAT_BITS_BIN)
 122                 return "-text";
 123         switch (convert_stats) {
 124         case CONVERT_STAT_BITS_TXT_LF:
 125                 return "lf";
 126         case CONVERT_STAT_BITS_TXT_CRLF:
 127                 return "crlf";
 128         case CONVERT_STAT_BITS_TXT_LF | CONVERT_STAT_BITS_TXT_CRLF:
 129                 return "mixed";
 130         default:
 131                 return "none";
 132         }
 133 }
 134
 135 const char *get_cached_convert_stats_ascii(const char *path)
 136 {
 137         const char *ret;
 138         unsigned long sz;
 139         void *data = read_blob_data_from_cache(path, &sz);
 140         ret = gather_convert_stats_ascii(data, sz);
 141         free(data);
 142         return ret;
 143 }
 144
 145 const char *get_wt_convert_stats_ascii(const char *path)
 146 {
 147         const char *ret = "";
 148         struct strbuf sb = STRBUF_INIT;
 149         if (strbuf_read_file(&sb, path, 0) >= 0)
 150                 ret = gather_convert_stats_ascii(sb.buf, sb.len);
 151         strbuf_release(&sb);
 152         return ret;
 153 }
 154
 155 static int text_eol_is_crlf(void)
 156 {
 157         if (auto_crlf == AUTO_CRLF_TRUE)
 158                 return 1;
 159         else if (auto_crlf == AUTO_CRLF_INPUT)
 160                 return 0;
 161         if (core_eol == EOL_CRLF)
 162                 return 1;
 163         if (core_eol == EOL_UNSET && EOL_NATIVE == EOL_CRLF)
 164                 return 1;
 165         return 0;
 166 }
 167
 168 static enum eol output_eol(enum crlf_action crlf_action)
 169 {
 170         switch (crlf_action) {
 171         case CRLF_BINARY:
 172                 return EOL_UNSET;
 173         case CRLF_TEXT_CRLF:
 174                 return EOL_CRLF;
 175         case CRLF_TEXT_INPUT:
 176                 return EOL_LF;
 177         case CRLF_UNDEFINED:
 178         case CRLF_AUTO_CRLF:
 179                 return EOL_CRLF;
 180         case CRLF_AUTO_INPUT:
 181                 return EOL_LF;
 182         case CRLF_TEXT:
 183         case CRLF_AUTO:
 184                 /* fall through */
 185                 return text_eol_is_crlf() ? EOL_CRLF : EOL_LF;
 186         }
 187         warning("Illegal crlf_action %d\n", (int)crlf_action);
 188         return core_eol;
 189 }
 190
 191 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
 192                             struct text_stat *stats, enum safe_crlf checksafe)
 193 {
 194         if (!checksafe)
 195                 return;
 196
 197         if (output_eol(crlf_action) == EOL_LF) {
 198                 /*
 199                  * CRLFs would not be restored by checkout:
 200                  * check if we'd remove CRLFs
 201                  */
 202                 if (stats->crlf) {
 203                         if (checksafe == SAFE_CRLF_WARN)
 204                                 warning("CRLF will be replaced by LF in %s.\nThe file will have its original line endings in your working directory.", path);
 205                         else /* i.e. SAFE_CRLF_FAIL */
 206                                 die("CRLF would be replaced by LF in %s.", path);
 207                 }
 208         } else if (output_eol(crlf_action) == EOL_CRLF) {
 209                 /*
 210                  * CRLFs would be added by checkout:
 211                  * check if we have "naked" LFs
 212                  */
 213                 if (stats->lonelf) {
 214                         if (checksafe == SAFE_CRLF_WARN)
 215                                 warning("LF will be replaced by CRLF in %s.\nThe file will have its original line endings in your working directory.", path);
 216                         else /* i.e. SAFE_CRLF_FAIL */
 217                                 die("LF would be replaced by CRLF in %s", path);
 218                 }
 219         }
 220 }
 221
 222 static int has_cr_in_index(const char *path)
 223 {
 224         unsigned long sz;
 225         void *data;
 226         int has_cr;
 227
 228         data = read_blob_data_from_cache(path, &sz);
 229         if (!data)
 230                 return 0;
 231         has_cr = memchr(data, '\r', sz) != NULL;
 232         free(data);
 233         return has_cr;
 234 }
 235
 236 static int crlf_to_git(const char *path, const char *src, size_t len,
 237                        struct strbuf *buf,
 238                        enum crlf_action crlf_action, enum safe_crlf checksafe)
 239 {
 240         struct text_stat stats;
 241         char *dst;
 242
 243         if (crlf_action == CRLF_BINARY ||
 244             (src && !len))
 245                 return 0;
 246
 247         /*
 248          * If we are doing a dry-run and have no source buffer, there is
 249          * nothing to analyze; we must assume we would convert.
 250          */
 251         if (!buf && !src)
 252                 return 1;
 253
 254         gather_stats(src, len, &stats);
 255
 256         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 257                 if (convert_is_binary(len, &stats))
 258                         return 0;
 259                 /*
 260                  * If the file in the index has any CR in it, do not convert.
 261                  * This is the new safer autocrlf handling.
 262                  */
 263                 if (checksafe == SAFE_CRLF_RENORMALIZE)
 264                         checksafe = SAFE_CRLF_FALSE;
 265                 else if (has_cr_in_index(path))
 266                         return 0;
 267         }
 268         check_safe_crlf(path, crlf_action, &stats, checksafe);
 269
 270         /* Optimization: No CRLF? Nothing to convert, regardless. */
 271         if (!stats.crlf)
 272                 return 0;
 273
 274         /*
 275          * At this point all of our source analysis is done, and we are sure we
 276          * would convert. If we are in dry-run mode, we can give an answer.
 277          */
 278         if (!buf)
 279                 return 1;
 280
 281         /* only grow if not in place */
 282         if (strbuf_avail(buf) + buf->len < len)
 283                 strbuf_grow(buf, len - buf->len);
 284         dst = buf->buf;
 285         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 286                 /*
 287                  * If we guessed, we already know we rejected a file with
 288                  * lone CR, and we can strip a CR without looking at what
 289                  * follow it.
 290                  */
 291                 do {
 292                         unsigned char c = *src++;
 293                         if (c != '\r')
 294                                 *dst++ = c;
 295                 } while (--len);
 296         } else {
 297                 do {
 298                         unsigned char c = *src++;
 299                         if (! (c == '\r' && (1 < len && *src == '\n')))
 300                                 *dst++ = c;
 301                 } while (--len);
 302         }
 303         strbuf_setlen(buf, dst - buf->buf);
 304         return 1;
 305 }
 306
 307 static int crlf_to_worktree(const char *path, const char *src, size_t len,
 308                             struct strbuf *buf, enum crlf_action crlf_action)
 309 {
 310         char *to_free = NULL;
 311         struct text_stat stats;
 312
 313         if (!len || output_eol(crlf_action) != EOL_CRLF)
 314                 return 0;
 315
 316         gather_stats(src, len, &stats);
 317
 318         /* No "naked" LF? Nothing to convert, regardless. */
 319         if (!stats.lonelf)
 320                 return 0;
 321
 322         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 323                 /* If we have any CR or CRLF line endings, we do not touch it */
 324                 /* This is the new safer autocrlf-handling */
 325                 if (stats.lonecr || stats.crlf )
 326                         return 0;
 327
 328                 if (convert_is_binary(len, &stats))
 329                         return 0;
 330         }
 331
 332         /* are we "faking" in place editing ? */
 333         if (src == buf->buf)
 334                 to_free = strbuf_detach(buf, NULL);
 335
 336         strbuf_grow(buf, len + stats.lonelf);
 337         for (;;) {
 338                 const char *nl = memchr(src, '\n', len);
 339                 if (!nl)
 340                         break;
 341                 if (nl > src && nl[-1] == '\r') {
 342                         strbuf_add(buf, src, nl + 1 - src);
 343                 } else {
 344                         strbuf_add(buf, src, nl - src);
 345                         strbuf_addstr(buf, "\r\n");
 346                 }
 347                 len -= nl + 1 - src;
 348                 src  = nl + 1;
 349         }
 350         strbuf_add(buf, src, len);
 351
 352         free(to_free);
 353         return 1;
 354 }
 355
 356 struct filter_params {
 357         const char *src;
 358         unsigned long size;
 359         int fd;
 360         const char *cmd;
 361         const char *path;
 362 };
 363
 364 static int filter_buffer_or_fd(int in, int out, void *data)
 365 {
 366         /*
 367          * Spawn cmd and feed the buffer contents through its stdin.
 368          */
 369         struct child_process child_process = CHILD_PROCESS_INIT;
 370         struct filter_params *params = (struct filter_params *)data;
 371         int write_err, status;
 372         const char *argv[] = { NULL, NULL };
 373
 374         /* apply % substitution to cmd */
 375         struct strbuf cmd = STRBUF_INIT;
 376         struct strbuf path = STRBUF_INIT;
 377         struct strbuf_expand_dict_entry dict[] = {
 378                 { "f", NULL, },
 379                 { NULL, NULL, },
 380         };
 381
 382         /* quote the path to preserve spaces, etc. */
 383         sq_quote_buf(&path, params->path);
 384         dict[0].value = path.buf;
 385
 386         /* expand all %f with the quoted path */
 387         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
 388         strbuf_release(&path);
 389
 390         argv[0] = cmd.buf;
 391
 392         child_process.argv = argv;
 393         child_process.use_shell = 1;
 394         child_process.in = -1;
 395         child_process.out = out;
 396
 397         if (start_command(&child_process))
 398                 return error("cannot fork to run external filter %s", params->cmd);
 399
 400         sigchain_push(SIGPIPE, SIG_IGN);
 401
 402         if (params->src) {
 403                 write_err = (write_in_full(child_process.in,
 404                                            params->src, params->size) < 0);
 405                 if (errno == EPIPE)
 406                         write_err = 0;
 407         } else {
 408                 write_err = copy_fd(params->fd, child_process.in);
 409                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
 410                         write_err = 0;
 411         }
 412
 413         if (close(child_process.in))
 414                 write_err = 1;
 415         if (write_err)
 416                 error("cannot feed the input to external filter %s", params->cmd);
 417
 418         sigchain_pop(SIGPIPE);
 419
 420         status = finish_command(&child_process);
 421         if (status)
 422                 error("external filter %s failed %d", params->cmd, status);
 423
 424         strbuf_release(&cmd);
 425         return (write_err || status);
 426 }
 427
 428 static int apply_filter(const char *path, const char *src, size_t len, int fd,
 429                         struct strbuf *dst, const char *cmd)
 430 {
 431         /*
 432          * Create a pipeline to have the command filter the buffer's
 433          * contents.
 434          *
 435          * (child --> cmd) --> us
 436          */
 437         int ret = 1;
 438         struct strbuf nbuf = STRBUF_INIT;
 439         struct async async;
 440         struct filter_params params;
 441
 442         if (!cmd || !*cmd)
 443                 return 0;
 444
 445         if (!dst)
 446                 return 1;
 447
 448         memset(&async, 0, sizeof(async));
 449         async.proc = filter_buffer_or_fd;
 450         async.data = &params;
 451         async.out = -1;
 452         params.src = src;
 453         params.size = len;
 454         params.fd = fd;
 455         params.cmd = cmd;
 456         params.path = path;
 457
 458         fflush(NULL);
 459         if (start_async(&async))
 460                 return 0;       /* error was already reported */
 461
 462         if (strbuf_read(&nbuf, async.out, len) < 0) {
 463                 error("read from external filter %s failed", cmd);
 464                 ret = 0;
 465         }
 466         if (close(async.out)) {
 467                 error("read from external filter %s failed", cmd);
 468                 ret = 0;
 469         }
 470         if (finish_async(&async)) {
 471                 error("external filter %s failed", cmd);
 472                 ret = 0;
 473         }
 474
 475         if (ret) {
 476                 strbuf_swap(dst, &nbuf);
 477         }
 478         strbuf_release(&nbuf);
 479         return ret;
 480 }
 481
 482 static struct convert_driver {
 483         const char *name;
 484         struct convert_driver *next;
 485         const char *smudge;
 486         const char *clean;
 487         int required;
 488 } *user_convert, **user_convert_tail;
 489
 490 static int read_convert_config(const char *var, const char *value, void *cb)
 491 {
 492         const char *key, *name;
 493         int namelen;
 494         struct convert_driver *drv;
 495
 496         /*
 497          * External conversion drivers are configured using
 498          * "filter.<name>.variable".
 499          */
 500         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
 501                 return 0;
 502         for (drv = user_convert; drv; drv = drv->next)
 503                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 504                         break;
 505         if (!drv) {
 506                 drv = xcalloc(1, sizeof(struct convert_driver));
 507                 drv->name = xmemdupz(name, namelen);
 508                 *user_convert_tail = drv;
 509                 user_convert_tail = &(drv->next);
 510         }
 511
 512         /*
 513          * filter.<name>.smudge and filter.<name>.clean specifies
 514          * the command line:
 515          *
 516          *      command-line
 517          *
 518          * The command-line will not be interpolated in any way.
 519          */
 520
 521         if (!strcmp("smudge", key))
 522                 return git_config_string(&drv->smudge, var, value);
 523
 524         if (!strcmp("clean", key))
 525                 return git_config_string(&drv->clean, var, value);
 526
 527         if (!strcmp("required", key)) {
 528                 drv->required = git_config_bool(var, value);
 529                 return 0;
 530         }
 531
 532         return 0;
 533 }
 534
 535 static int count_ident(const char *cp, unsigned long size)
 536 {
 537         /*
 538          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 539          */
 540         int cnt = 0;
 541         char ch;
 542
 543         while (size) {
 544                 ch = *cp++;
 545                 size--;
 546                 if (ch != '$')
 547                         continue;
 548                 if (size < 3)
 549                         break;
 550                 if (memcmp("Id", cp, 2))
 551                         continue;
 552                 ch = cp[2];
 553                 cp += 3;
 554                 size -= 3;
 555                 if (ch == '$')
 556                         cnt++; /* $Id$ */
 557                 if (ch != ':')
 558                         continue;
 559
 560                 /*
 561                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 562                  */
 563                 while (size) {
 564                         ch = *cp++;
 565                         size--;
 566                         if (ch == '$') {
 567                                 cnt++;
 568                                 break;
 569                         }
 570                         if (ch == '\n')
 571                                 break;
 572                 }
 573         }
 574         return cnt;
 575 }
 576
 577 static int ident_to_git(const char *path, const char *src, size_t len,
 578                         struct strbuf *buf, int ident)
 579 {
 580         char *dst, *dollar;
 581
 582         if (!ident || (src && !count_ident(src, len)))
 583                 return 0;
 584
 585         if (!buf)
 586                 return 1;
 587
 588         /* only grow if not in place */
 589         if (strbuf_avail(buf) + buf->len < len)
 590                 strbuf_grow(buf, len - buf->len);
 591         dst = buf->buf;
 592         for (;;) {
 593                 dollar = memchr(src, '$', len);
 594                 if (!dollar)
 595                         break;
 596                 memmove(dst, src, dollar + 1 - src);
 597                 dst += dollar + 1 - src;
 598                 len -= dollar + 1 - src;
 599                 src  = dollar + 1;
 600
 601                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 602                         dollar = memchr(src + 3, '$', len - 3);
 603                         if (!dollar)
 604                                 break;
 605                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 606                                 /* Line break before the next dollar. */
 607                                 continue;
 608                         }
 609
 610                         memcpy(dst, "Id$", 3);
 611                         dst += 3;
 612                         len -= dollar + 1 - src;
 613                         src  = dollar + 1;
 614                 }
 615         }
 616         memmove(dst, src, len);
 617         strbuf_setlen(buf, dst + len - buf->buf);
 618         return 1;
 619 }
 620
 621 static int ident_to_worktree(const char *path, const char *src, size_t len,
 622                              struct strbuf *buf, int ident)
 623 {
 624         unsigned char sha1[20];
 625         char *to_free = NULL, *dollar, *spc;
 626         int cnt;
 627
 628         if (!ident)
 629                 return 0;
 630
 631         cnt = count_ident(src, len);
 632         if (!cnt)
 633                 return 0;
 634
 635         /* are we "faking" in place editing ? */
 636         if (src == buf->buf)
 637                 to_free = strbuf_detach(buf, NULL);
 638         hash_sha1_file(src, len, "blob", sha1);
 639
 640         strbuf_grow(buf, len + cnt * 43);
 641         for (;;) {
 642                 /* step 1: run to the next '$' */
 643                 dollar = memchr(src, '$', len);
 644                 if (!dollar)
 645                         break;
 646                 strbuf_add(buf, src, dollar + 1 - src);
 647                 len -= dollar + 1 - src;
 648                 src  = dollar + 1;
 649
 650                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 651                 if (len < 3 || memcmp("Id", src, 2))
 652                         continue;
 653
 654                 /* step 3: skip over Id$ or Id:xxxxx$ */
 655                 if (src[2] == '$') {
 656                         src += 3;
 657                         len -= 3;
 658                 } else if (src[2] == ':') {
 659                         /*
 660                          * It's possible that an expanded Id has crept its way into the
 661                          * repository, we cope with that by stripping the expansion out.
 662                          * This is probably not a good idea, since it will cause changes
 663                          * on checkout, which won't go away by stash, but let's keep it
 664                          * for git-style ids.
 665                          */
 666                         dollar = memchr(src + 3, '$', len - 3);
 667                         if (!dollar) {
 668                                 /* incomplete keyword, no more '$', so just quit the loop */
 669                                 break;
 670                         }
 671
 672                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 673                                 /* Line break before the next dollar. */
 674                                 continue;
 675                         }
 676
 677                         spc = memchr(src + 4, ' ', dollar - src - 4);
 678                         if (spc && spc < dollar-1) {
 679                                 /* There are spaces in unexpected places.
 680                                  * This is probably an id from some other
 681                                  * versioning system. Keep it for now.
 682                                  */
 683                                 continue;
 684                         }
 685
 686                         len -= dollar + 1 - src;
 687                         src  = dollar + 1;
 688                 } else {
 689                         /* it wasn't a "Id$" or "Id:xxxx$" */
 690                         continue;
 691                 }
 692
 693                 /* step 4: substitute */
 694                 strbuf_addstr(buf, "Id: ");
 695                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 696                 strbuf_addstr(buf, " $");
 697         }
 698         strbuf_add(buf, src, len);
 699
 700         free(to_free);
 701         return 1;
 702 }
 703
 704 static enum crlf_action git_path_check_crlf(struct git_attr_check *check)
 705 {
 706         const char *value = check->value;
 707
 708         if (ATTR_TRUE(value))
 709                 return CRLF_TEXT;
 710         else if (ATTR_FALSE(value))
 711                 return CRLF_BINARY;
 712         else if (ATTR_UNSET(value))
 713                 ;
 714         else if (!strcmp(value, "input"))
 715                 return CRLF_TEXT_INPUT;
 716         else if (!strcmp(value, "auto"))
 717                 return CRLF_AUTO;
 718         return CRLF_UNDEFINED;
 719 }
 720
 721 static enum eol git_path_check_eol(struct git_attr_check *check)
 722 {
 723         const char *value = check->value;
 724
 725         if (ATTR_UNSET(value))
 726                 ;
 727         else if (!strcmp(value, "lf"))
 728                 return EOL_LF;
 729         else if (!strcmp(value, "crlf"))
 730                 return EOL_CRLF;
 731         return EOL_UNSET;
 732 }
 733
 734 static struct convert_driver *git_path_check_convert(struct git_attr_check *check)
 735 {
 736         const char *value = check->value;
 737         struct convert_driver *drv;
 738
 739         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
 740                 return NULL;
 741         for (drv = user_convert; drv; drv = drv->next)
 742                 if (!strcmp(value, drv->name))
 743                         return drv;
 744         return NULL;
 745 }
 746
 747 static int git_path_check_ident(struct git_attr_check *check)
 748 {
 749         const char *value = check->value;
 750
 751         return !!ATTR_TRUE(value);
 752 }
 753
 754 struct conv_attrs {
 755         struct convert_driver *drv;
 756         enum crlf_action attr_action; /* What attr says */
 757         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
 758         int ident;
 759 };
 760
 761 static const char *conv_attr_name[] = {
 762         "crlf", "ident", "filter", "eol", "text",
 763 };
 764 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
 765
 766 static void convert_attrs(struct conv_attrs *ca, const char *path)
 767 {
 768         int i;
 769         static struct git_attr_check ccheck[NUM_CONV_ATTRS];
 770
 771         if (!ccheck[0].attr) {
 772                 for (i = 0; i < NUM_CONV_ATTRS; i++)
 773                         ccheck[i].attr = git_attr(conv_attr_name[i]);
 774                 user_convert_tail = &user_convert;
 775                 git_config(read_convert_config, NULL);
 776         }
 777
 778         if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
 779                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
 780                 if (ca->crlf_action == CRLF_UNDEFINED)
 781                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
 782                 ca->attr_action = ca->crlf_action;
 783                 ca->ident = git_path_check_ident(ccheck + 1);
 784                 ca->drv = git_path_check_convert(ccheck + 2);
 785                 if (ca->crlf_action != CRLF_BINARY) {
 786                         enum eol eol_attr = git_path_check_eol(ccheck + 3);
 787                         if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
 788                                 ca->crlf_action = CRLF_AUTO_INPUT;
 789                         else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
 790                                 ca->crlf_action = CRLF_AUTO_CRLF;
 791                         else if (eol_attr == EOL_LF)
 792                                 ca->crlf_action = CRLF_TEXT_INPUT;
 793                         else if (eol_attr == EOL_CRLF)
 794                                 ca->crlf_action = CRLF_TEXT_CRLF;
 795                 }
 796                 ca->attr_action = ca->crlf_action;
 797         } else {
 798                 ca->drv = NULL;
 799                 ca->crlf_action = CRLF_UNDEFINED;
 800                 ca->ident = 0;
 801         }
 802         if (ca->crlf_action == CRLF_TEXT)
 803                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
 804         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
 805                 ca->crlf_action = CRLF_BINARY;
 806         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
 807                 ca->crlf_action = CRLF_AUTO_CRLF;
 808         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
 809                 ca->crlf_action = CRLF_AUTO_INPUT;
 810 }
 811
 812 int would_convert_to_git_filter_fd(const char *path)
 813 {
 814         struct conv_attrs ca;
 815
 816         convert_attrs(&ca, path);
 817         if (!ca.drv)
 818                 return 0;
 819
 820         /*
 821          * Apply a filter to an fd only if the filter is required to succeed.
 822          * We must die if the filter fails, because the original data before
 823          * filtering is not available.
 824          */
 825         if (!ca.drv->required)
 826                 return 0;
 827
 828         return apply_filter(path, NULL, 0, -1, NULL, ca.drv->clean);
 829 }
 830
 831 const char *get_convert_attr_ascii(const char *path)
 832 {
 833         struct conv_attrs ca;
 834
 835         convert_attrs(&ca, path);
 836         switch (ca.attr_action) {
 837         case CRLF_UNDEFINED:
 838                 return "";
 839         case CRLF_BINARY:
 840                 return "-text";
 841         case CRLF_TEXT:
 842                 return "text";
 843         case CRLF_TEXT_INPUT:
 844                 return "text eol=lf";
 845         case CRLF_TEXT_CRLF:
 846                 return "text eol=crlf";
 847         case CRLF_AUTO:
 848                 return "text=auto";
 849         case CRLF_AUTO_CRLF:
 850                 return "text=auto eol=crlf";
 851         case CRLF_AUTO_INPUT:
 852                 return "text=auto eol=lf";
 853         }
 854         return "";
 855 }
 856
 857 int convert_to_git(const char *path, const char *src, size_t len,
 858                    struct strbuf *dst, enum safe_crlf checksafe)
 859 {
 860         int ret = 0;
 861         const char *filter = NULL;
 862         int required = 0;
 863         struct conv_attrs ca;
 864
 865         convert_attrs(&ca, path);
 866         if (ca.drv) {
 867                 filter = ca.drv->clean;
 868                 required = ca.drv->required;
 869         }
 870
 871         ret |= apply_filter(path, src, len, -1, dst, filter);
 872         if (!ret && required)
 873                 die("%s: clean filter '%s' failed", path, ca.drv->name);
 874
 875         if (ret && dst) {
 876                 src = dst->buf;
 877                 len = dst->len;
 878         }
 879         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
 880         if (ret && dst) {
 881                 src = dst->buf;
 882                 len = dst->len;
 883         }
 884         return ret | ident_to_git(path, src, len, dst, ca.ident);
 885 }
 886
 887 void convert_to_git_filter_fd(const char *path, int fd, struct strbuf *dst,
 888                               enum safe_crlf checksafe)
 889 {
 890         struct conv_attrs ca;
 891         convert_attrs(&ca, path);
 892
 893         assert(ca.drv);
 894         assert(ca.drv->clean);
 895
 896         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv->clean))
 897                 die("%s: clean filter '%s' failed", path, ca.drv->name);
 898
 899         crlf_to_git(path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
 900         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
 901 }
 902
 903 static int convert_to_working_tree_internal(const char *path, const char *src,
 904                                             size_t len, struct strbuf *dst,
 905                                             int normalizing)
 906 {
 907         int ret = 0, ret_filter = 0;
 908         const char *filter = NULL;
 909         int required = 0;
 910         struct conv_attrs ca;
 911
 912         convert_attrs(&ca, path);
 913         if (ca.drv) {
 914                 filter = ca.drv->smudge;
 915                 required = ca.drv->required;
 916         }
 917
 918         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
 919         if (ret) {
 920                 src = dst->buf;
 921                 len = dst->len;
 922         }
 923         /*
 924          * CRLF conversion can be skipped if normalizing, unless there
 925          * is a smudge filter.  The filter might expect CRLFs.
 926          */
 927         if (filter || !normalizing) {
 928                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
 929                 if (ret) {
 930                         src = dst->buf;
 931                         len = dst->len;
 932                 }
 933         }
 934
 935         ret_filter = apply_filter(path, src, len, -1, dst, filter);
 936         if (!ret_filter && required)
 937                 die("%s: smudge filter %s failed", path, ca.drv->name);
 938
 939         return ret | ret_filter;
 940 }
 941
 942 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
 943 {
 944         return convert_to_working_tree_internal(path, src, len, dst, 0);
 945 }
 946
 947 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
 948 {
 949         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
 950         if (ret) {
 951                 src = dst->buf;
 952                 len = dst->len;
 953         }
 954         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_RENORMALIZE);
 955 }
 956
 957 /*****************************************************************
 958  *
 959  * Streaming conversion support
 960  *
 961  *****************************************************************/
 962
 963 typedef int (*filter_fn)(struct stream_filter *,
 964                          const char *input, size_t *isize_p,
 965                          char *output, size_t *osize_p);
 966 typedef void (*free_fn)(struct stream_filter *);
 967
 968 struct stream_filter_vtbl {
 969         filter_fn filter;
 970         free_fn free;
 971 };
 972
 973 struct stream_filter {
 974         struct stream_filter_vtbl *vtbl;
 975 };
 976
 977 static int null_filter_fn(struct stream_filter *filter,
 978                           const char *input, size_t *isize_p,
 979                           char *output, size_t *osize_p)
 980 {
 981         size_t count;
 982
 983         if (!input)
 984                 return 0; /* we do not keep any states */
 985         count = *isize_p;
 986         if (*osize_p < count)
 987                 count = *osize_p;
 988         if (count) {
 989                 memmove(output, input, count);
 990                 *isize_p -= count;
 991                 *osize_p -= count;
 992         }
 993         return 0;
 994 }
 995
 996 static void null_free_fn(struct stream_filter *filter)
 997 {
 998         ; /* nothing -- null instances are shared */
 999 }
1000
1001 static struct stream_filter_vtbl null_vtbl = {
1002         null_filter_fn,
1003         null_free_fn,
1004 };
1005
1006 static struct stream_filter null_filter_singleton = {
1007         &null_vtbl,
1008 };
1009
1010 int is_null_stream_filter(struct stream_filter *filter)
1011 {
1012         return filter == &null_filter_singleton;
1013 }
1014
1015
1016 /*
1017  * LF-to-CRLF filter
1018  */
1019
1020 struct lf_to_crlf_filter {
1021         struct stream_filter filter;
1022         unsigned has_held:1;
1023         char held;
1024 };
1025
1026 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1027                                 const char *input, size_t *isize_p,
1028                                 char *output, size_t *osize_p)
1029 {
1030         size_t count, o = 0;
1031         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1032
1033         /*
1034          * We may be holding onto the CR to see if it is followed by a
1035          * LF, in which case we would need to go to the main loop.
1036          * Otherwise, just emit it to the output stream.
1037          */
1038         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1039                 output[o++] = lf_to_crlf->held;
1040                 lf_to_crlf->has_held = 0;
1041         }
1042
1043         /* We are told to drain */
1044         if (!input) {
1045                 *osize_p -= o;
1046                 return 0;
1047         }
1048
1049         count = *isize_p;
1050         if (count || lf_to_crlf->has_held) {
1051                 size_t i;
1052                 int was_cr = 0;
1053
1054                 if (lf_to_crlf->has_held) {
1055                         was_cr = 1;
1056                         lf_to_crlf->has_held = 0;
1057                 }
1058
1059                 for (i = 0; o < *osize_p && i < count; i++) {
1060                         char ch = input[i];
1061
1062                         if (ch == '\n') {
1063                                 output[o++] = '\r';
1064                         } else if (was_cr) {
1065                                 /*
1066                                  * Previous round saw CR and it is not followed
1067                                  * by a LF; emit the CR before processing the
1068                                  * current character.
1069                                  */
1070                                 output[o++] = '\r';
1071                         }
1072
1073                         /*
1074                          * We may have consumed the last output slot,
1075                          * in which case we need to break out of this
1076                          * loop; hold the current character before
1077                          * returning.
1078                          */
1079                         if (*osize_p <= o) {
1080                                 lf_to_crlf->has_held = 1;
1081                                 lf_to_crlf->held = ch;
1082                                 continue; /* break but increment i */
1083                         }
1084
1085                         if (ch == '\r') {
1086                                 was_cr = 1;
1087                                 continue;
1088                         }
1089
1090                         was_cr = 0;
1091                         output[o++] = ch;
1092                 }
1093
1094                 *osize_p -= o;
1095                 *isize_p -= i;
1096
1097                 if (!lf_to_crlf->has_held && was_cr) {
1098                         lf_to_crlf->has_held = 1;
1099                         lf_to_crlf->held = '\r';
1100                 }
1101         }
1102         return 0;
1103 }
1104
1105 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1106 {
1107         free(filter);
1108 }
1109
1110 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1111         lf_to_crlf_filter_fn,
1112         lf_to_crlf_free_fn,
1113 };
1114
1115 static struct stream_filter *lf_to_crlf_filter(void)
1116 {
1117         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1118
1119         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1120         return (struct stream_filter *)lf_to_crlf;
1121 }
1122
1123 /*
1124  * Cascade filter
1125  */
1126 #define FILTER_BUFFER 1024
1127 struct cascade_filter {
1128         struct stream_filter filter;
1129         struct stream_filter *one;
1130         struct stream_filter *two;
1131         char buf[FILTER_BUFFER];
1132         int end, ptr;
1133 };
1134
1135 static int cascade_filter_fn(struct stream_filter *filter,
1136                              const char *input, size_t *isize_p,
1137                              char *output, size_t *osize_p)
1138 {
1139         struct cascade_filter *cas = (struct cascade_filter *) filter;
1140         size_t filled = 0;
1141         size_t sz = *osize_p;
1142         size_t to_feed, remaining;
1143
1144         /*
1145          * input -- (one) --> buf -- (two) --> output
1146          */
1147         while (filled < sz) {
1148                 remaining = sz - filled;
1149
1150                 /* do we already have something to feed two with? */
1151                 if (cas->ptr < cas->end) {
1152                         to_feed = cas->end - cas->ptr;
1153                         if (stream_filter(cas->two,
1154                                           cas->buf + cas->ptr, &to_feed,
1155                                           output + filled, &remaining))
1156                                 return -1;
1157                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1158                         filled = sz - remaining;
1159                         continue;
1160                 }
1161
1162                 /* feed one from upstream and have it emit into our buffer */
1163                 to_feed = input ? *isize_p : 0;
1164                 if (input && !to_feed)
1165                         break;
1166                 remaining = sizeof(cas->buf);
1167                 if (stream_filter(cas->one,
1168                                   input, &to_feed,
1169                                   cas->buf, &remaining))
1170                         return -1;
1171                 cas->end = sizeof(cas->buf) - remaining;
1172                 cas->ptr = 0;
1173                 if (input) {
1174                         size_t fed = *isize_p - to_feed;
1175                         *isize_p -= fed;
1176                         input += fed;
1177                 }
1178
1179                 /* do we know that we drained one completely? */
1180                 if (input || cas->end)
1181                         continue;
1182
1183                 /* tell two to drain; we have nothing more to give it */
1184                 to_feed = 0;
1185                 remaining = sz - filled;
1186                 if (stream_filter(cas->two,
1187                                   NULL, &to_feed,
1188                                   output + filled, &remaining))
1189                         return -1;
1190                 if (remaining == (sz - filled))
1191                         break; /* completely drained two */
1192                 filled = sz - remaining;
1193         }
1194         *osize_p -= filled;
1195         return 0;
1196 }
1197
1198 static void cascade_free_fn(struct stream_filter *filter)
1199 {
1200         struct cascade_filter *cas = (struct cascade_filter *)filter;
1201         free_stream_filter(cas->one);
1202         free_stream_filter(cas->two);
1203         free(filter);
1204 }
1205
1206 static struct stream_filter_vtbl cascade_vtbl = {
1207         cascade_filter_fn,
1208         cascade_free_fn,
1209 };
1210
1211 static struct stream_filter *cascade_filter(struct stream_filter *one,
1212                                             struct stream_filter *two)
1213 {
1214         struct cascade_filter *cascade;
1215
1216         if (!one || is_null_stream_filter(one))
1217                 return two;
1218         if (!two || is_null_stream_filter(two))
1219                 return one;
1220
1221         cascade = xmalloc(sizeof(*cascade));
1222         cascade->one = one;
1223         cascade->two = two;
1224         cascade->end = cascade->ptr = 0;
1225         cascade->filter.vtbl = &cascade_vtbl;
1226         return (struct stream_filter *)cascade;
1227 }
1228
1229 /*
1230  * ident filter
1231  */
1232 #define IDENT_DRAINING (-1)
1233 #define IDENT_SKIPPING (-2)
1234 struct ident_filter {
1235         struct stream_filter filter;
1236         struct strbuf left;
1237         int state;
1238         char ident[45]; /* ": x40 $" */
1239 };
1240
1241 static int is_foreign_ident(const char *str)
1242 {
1243         int i;
1244
1245         if (!skip_prefix(str, "$Id: ", &str))
1246                 return 0;
1247         for (i = 0; str[i]; i++) {
1248                 if (isspace(str[i]) && str[i+1] != '$')
1249                         return 1;
1250         }
1251         return 0;
1252 }
1253
1254 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1255 {
1256         size_t to_drain = ident->left.len;
1257
1258         if (*osize_p < to_drain)
1259                 to_drain = *osize_p;
1260         if (to_drain) {
1261                 memcpy(*output_p, ident->left.buf, to_drain);
1262                 strbuf_remove(&ident->left, 0, to_drain);
1263                 *output_p += to_drain;
1264                 *osize_p -= to_drain;
1265         }
1266         if (!ident->left.len)
1267                 ident->state = 0;
1268 }
1269
1270 static int ident_filter_fn(struct stream_filter *filter,
1271                            const char *input, size_t *isize_p,
1272                            char *output, size_t *osize_p)
1273 {
1274         struct ident_filter *ident = (struct ident_filter *)filter;
1275         static const char head[] = "$Id";
1276
1277         if (!input) {
1278                 /* drain upon eof */
1279                 switch (ident->state) {
1280                 default:
1281                         strbuf_add(&ident->left, head, ident->state);
1282                 case IDENT_SKIPPING:
1283                         /* fallthru */
1284                 case IDENT_DRAINING:
1285                         ident_drain(ident, &output, osize_p);
1286                 }
1287                 return 0;
1288         }
1289
1290         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1291                 int ch;
1292
1293                 if (ident->state == IDENT_DRAINING) {
1294                         ident_drain(ident, &output, osize_p);
1295                         if (!*osize_p)
1296                                 break;
1297                         continue;
1298                 }
1299
1300                 ch = *(input++);
1301                 (*isize_p)--;
1302
1303                 if (ident->state == IDENT_SKIPPING) {
1304                         /*
1305                          * Skipping until '$' or LF, but keeping them
1306                          * in case it is a foreign ident.
1307                          */
1308                         strbuf_addch(&ident->left, ch);
1309                         if (ch != '\n' && ch != '$')
1310                                 continue;
1311                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1312                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1313                                 strbuf_addstr(&ident->left, ident->ident);
1314                         }
1315                         ident->state = IDENT_DRAINING;
1316                         continue;
1317                 }
1318
1319                 if (ident->state < sizeof(head) &&
1320                     head[ident->state] == ch) {
1321                         ident->state++;
1322                         continue;
1323                 }
1324
1325                 if (ident->state)
1326                         strbuf_add(&ident->left, head, ident->state);
1327                 if (ident->state == sizeof(head) - 1) {
1328                         if (ch != ':' && ch != '$') {
1329                                 strbuf_addch(&ident->left, ch);
1330                                 ident->state = 0;
1331                                 continue;
1332                         }
1333
1334                         if (ch == ':') {
1335                                 strbuf_addch(&ident->left, ch);
1336                                 ident->state = IDENT_SKIPPING;
1337                         } else {
1338                                 strbuf_addstr(&ident->left, ident->ident);
1339                                 ident->state = IDENT_DRAINING;
1340                         }
1341                         continue;
1342                 }
1343
1344                 strbuf_addch(&ident->left, ch);
1345                 ident->state = IDENT_DRAINING;
1346         }
1347         return 0;
1348 }
1349
1350 static void ident_free_fn(struct stream_filter *filter)
1351 {
1352         struct ident_filter *ident = (struct ident_filter *)filter;
1353         strbuf_release(&ident->left);
1354         free(filter);
1355 }
1356
1357 static struct stream_filter_vtbl ident_vtbl = {
1358         ident_filter_fn,
1359         ident_free_fn,
1360 };
1361
1362 static struct stream_filter *ident_filter(const unsigned char *sha1)
1363 {
1364         struct ident_filter *ident = xmalloc(sizeof(*ident));
1365
1366         xsnprintf(ident->ident, sizeof(ident->ident),
1367                   ": %s $", sha1_to_hex(sha1));
1368         strbuf_init(&ident->left, 0);
1369         ident->filter.vtbl = &ident_vtbl;
1370         ident->state = 0;
1371         return (struct stream_filter *)ident;
1372 }
1373
1374 /*
1375  * Return an appropriately constructed filter for the path, or NULL if
1376  * the contents cannot be filtered without reading the whole thing
1377  * in-core.
1378  *
1379  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1380  * large binary blob you would want us not to slurp into the memory!
1381  */
1382 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1383 {
1384         struct conv_attrs ca;
1385         struct stream_filter *filter = NULL;
1386
1387         convert_attrs(&ca, path);
1388         if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1389                 return NULL;
1390
1391         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1392                 return NULL;
1393
1394         if (ca.ident)
1395                 filter = ident_filter(sha1);
1396
1397         if (output_eol(ca.crlf_action) == EOL_CRLF)
1398                 filter = cascade_filter(filter, lf_to_crlf_filter());
1399         else
1400                 filter = cascade_filter(filter, &null_filter_singleton);
1401
1402         return filter;
1403 }
1404
1405 void free_stream_filter(struct stream_filter *filter)
1406 {
1407         filter->vtbl->free(filter);
1408 }
1409
1410 int stream_filter(struct stream_filter *filter,
1411                   const char *input, size_t *isize_p,
1412                   char *output, size_t *osize_p)
1413 {
1414         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1415 }